Iterative Least Squares Detection using the EM-algorithm
Patrik Bohlin · 2002
In this paper an algorithm based on Expectation-Maximization (EM) is derived for the problem of separating superimposed digitally modulated signals impinging on an antenna array. It is found that this algorithm closely resembles previously proposed methods based on Iterative Least Squares (ILS) techniques. Using an extension to the EM-algorithm known as SAGE, improvements to the algorithms are proposed for increasing performance and convergence rate as well as handling unknown noise covariance matrices. These improvements can in the ILS-framework be seen as doing Gauss-Seidel optimization instead of ¯x-point optimization as is done in the original algorithm.